The occurrence statistics of equatorial plasma bubbles (EPBs) obtained from low-inclination orbit satellites are significantly affected by the way the data are sampled and the way that the EPBs are counted. To resolve the discrepancy between the EPB occurrence frequency determined by ground-based observations and in situ sampling of plasma density from spacecraft, we have developed a new EPB detection method that minimizes the dependence of the EPB occurrence rate on the data processing method. The global EPB distribution maps are created by analyzing the measurements of the ion density from the first Republic of China satellite (ROCSAT-1) during March 1999 to June 2004. The EPB occurrence probability obtained using our new EPB detection method is a few times greater than that obtained using the conventional method. Our results are comparable to the ground observations. The good agreement of the global EPB distribution with the global morphology of the evening prereversal enhancement (PRE) of vertical ion velocity supports the notion that the PRE is an important factor on a global scale in the generation of EPBs. However, the generation of EPBs is not guaranteed by the occurrence of an intense PRE. Other mechanisms, in addition to the PRE, should be considered as an explanation for the occurrence of EPBs on the topside.
 Severe ionospheric turbulence at low latitudes is primarily manifested by the phenomenon known as equatorial plasma bubble (EPB) or equatorial spread F (ESF). The rapid loss of molecular ions after sunset by recombination produces a steep vertical plasma density gradient in the bottomside F region and the ionosphere becomes unstable to the gravitational Rayleigh-Taylor (R-T) instability. The polarization electric field induced by the modulation of the bottomside drives the upward drift of low-density plasma in the bottomside much like the rise of air bubbles in a liquid. EPBs develop suddenly, just after sunset, and then decay as time progresses. As a result, the local time distribution of the EPB occurrence probability shows a rapid increase in the probability of occurrence before 2000 LT and a gradual decrease after 2000 LT [e.g., Su et al., 2006].
Anderson et al.  investigated the correlation between the intensity of UHF S4 index and the PRE in the Peruvian/Chilean longitude sector during 1998–1999. Their results showed the occurrence of scintillation (S4 > 0.5) about 90% of the time when the PRE was greater than 20 m s−1. They posited that the seeding mechanism is always present and that the necessary and sufficient condition for generation of scintillation is a PRE greater than some critical value. Basu et al.  also suggested that an upward velocity of 10∼20 m s−1 is necessary for the commencement of ESF in the Peruvian sector. On the other hand, the ROCSAT-1 observations show that the EPB occurrence probability is less than 40% even for PRE values greater than 40 m s−1 [Li et al., 2008; Su et al., 2008].
 The occurrence probability of the electron density irregularities can be different from the occurrence probability of scintillation or ESF since different quantities are measured using different criteria during different periods. However, the major reason that the reported occurrence probability is small for a low-inclination orbit satellite is because of way the occurrence is calculated: the calculation of the occurrence probability is dependent on the way the data are sampled and the way that the EPBs are counted. Remember, on any given low-inclination orbit a very large range of local solar time is sampled. These problems are less pronounced in the polar orbit satellite observations since the polar orbit satellites sample the equatorial region at a fixed local time. We note that the EPB occurrence probability obtained from ROCSAT-1 at 600 km [Su et al., 2006] is even smaller than that obtained from the Defense Meteorological Satellite Program (DMSP) at 840 km during the same period [Burke et al., 2004a; Gentile et al., 2006]. Errors in the EPB occurrence statistics affect the forecasting capability of EPBs. People who develop the EPB forecasting model based on the ground observation data would predict occurrence of EPBs a few times higher than people who use the low-inclination orbit satellite data. Uncertainty in the EPB occurrence statistics also affects the verification of the onset conditions of EPBs.
 We have developed a new EPB detection method that may minimize the dependence of the EPB occurrence probability on the data sampling and EPB counting methods. Global EPB distribution maps are produced by applying our new EPB detection method to the ROCSAT-1 ion density data during March 1999 to June 2004. Using these new EPB occurrence statistics, we investigate the seasonal longitudinal variability of the EPB occurrence probability and its association with the PRE.
2. A New EPB Detection Method
 ROCSAT-1 was in a circular orbit at 600 km with an orbital inclination of 35° and provided data from March 1999 to June 2004. Its low-inclination orbit enabled ROCSAT-1 to sample the ionosphere at all local times. The Ionospheric Plasma and Electrodynamics Instrument (IPEI) on board the ROCSAT-1 consists of four sensors that measure the ion concentration, ion temperature, and ion drift velocities [Su et al., 1999; Yeh et al., 1999]. The IPEI was operated with a 100% duty cycle. In this study, we analyze the ion density measurements to produce a global EPB distribution map. The mean vertical ion velocities at 1730–1930 LT within ±5° dip latitude are used as representative of the PRE.
 Typically, EPBs in satellite data are detected using standard deviations of ion density during a certain time segment. Su et al.  used 10-s ROCSAT-1 data to calculate a parameter σ defined as
Here ni is the ion density at each second and noi is the linearly fitted value. Kil and Heelis  calculated σ similarly using the 8-s segment linear-scale ion density data from the AE-E satellite. Su et al.  used σ = 0.3% to identify the locations of EPBs. Once the EPB locations are identified, the EPB occurrence rate is calculated using the ratio of the number of EPB data points to the total data points. They calculated the mean EPB occurrence probability using the ROCSAT-1 data between 1800 and 0600 LT within ±15° dip latitudes. As one can imagine, the mean EPB occurrence probability calculated using the conventional method (described by Su et al. ) varies with the selection of the ranges of the local time and latitudes. For example, the mean EPB occurrence probability calculated using data between 2000 and 0000 LT would be greater than that calculated using data between 1800 and 0600 LT since the peak of the EPB occurrence probability occurs near 2000 LT. EPBs are rarely detected by satellites before 1900 LT. Inclusion of data between 1800 and 1900 LT will naturally lower the mean EPB occurrence probability. The EPB occurrence probability has its peak at the dip equator [Kil and Heelis, 1998; Su et al., 2006]. Therefore, the mean EPB occurrence probability calculated using data within ±10° dip latitudes would be greater than that calculated using data within ±20° dip latitudes.
 Our new EPB detection method defines the daily occurrence of EPBs in three categories: (1) “bubble present,” (0) “bubble absent,” and (2) “undetermined.” If EPBs were detected between 1900 and 0600 LT within ±20° dip latitudes on the given night at the given longitude bin, this longitude bin is recorded with “1”. If EPBs were not detected on that night, it could be either because EPBs were not created on that night or because the data sampled local times or latitudes were not suitable for the detection of EPBs. If the satellite samples data around 0400 LT and 15° dip latitude at a certain longitude, for example, the satellite may not detect EPBs, although EPBs were created near the dip equator before midnight at that longitude. We use the occurrence of the satellite orbit within ±10° dip latitudes between 2000 and 0000 LT to determine the indices “0” or “2”. Selection of these ranges is not unique. Widening of these ranges would reduce the “undetermined” portion in the daily EPB occurrence map but would increase the uncertainty in the EPB occurrence probability. Figure 1 schematically explains the determination method of the EPB occurrence on each day in a given longitude bin. The diagram shows the daily variation of the satellite orbit in local time (1900–0600 LT) and dip latitude (20°S–20°N) coordinates. The thin lines indicate the satellite orbits. Detection of EPBs along satellite paths is indicated by thick lines. The small box indicates the region between 2000 and 0000 LT within ±10° dip latitudes. On Day 1, the occurrence of EPBs is not determined (index 2) since the satellite does not have a path within the box. The absence of EPBs on Day 1 is either because the data were sampled at too high a latitude or because EPBs were not generated on that night. On Day 2, the satellite orbit had a segment inside the box, but EPBs were not detected. So, we record the absence of EPBs (index 0) on Day 2. EPBs were detected on Day 3 and Day 4, and our data set records the occurrence of EPBs (index 1) on those days. An EPB detection will be counted at a given longitude even if the satellite is outside of the latitude and local time box. On Day 5, the satellite sampled data too late in local time (the satellite path is out of the box) and the occurrence of EPBs is not determined (index 2).
 We use a parameter similar to σ defined in equation (1) to find the locations of EPBs. In our new method, we use 100-s data instead of 10-s data. The background density is then represented using an 11-point smoothing curve. Our tests showed that the EPB locations could be determined more accurately after detrending the data by using the multipoint smoothing curve rather than by representing the background by a linear fit. Our new (Figures 2c–2f) and conventional (Figures 2g–2j) EPB detection methods are applied to the ROCSAT-1 data on 22 September 2002 (Figures 2a and 2b). Figures 2c and 2d show σ values calculated using the new method. Figures 2e and 2f show the EPB occurrence probability identified with σ greater than 0.3%. Our new method identifies the occurrence of EPBs continuously in broad longitude regions. Figures 2g and 2h show σ values calculated using 10-s data and Figures 2i and 2j show the corresponding percentage EPB occurrence calculated using the same σ criterion. The EPB occurrence probability calculated using the conventional method (Figures 2i and 2j) shows large fluctuations in the region where the new EPB detection method identifies 100% occurrence of EPBs. Actually, the exact locations of EPBs at the particular spacecraft altitude are better represented in the conventional method than in the new method. The question is which method better represents the occurrence of ionospheric turbulence in a region. EPBs extend a few to several hundred kilometers in altitude. EPBs are also tilted altitudinally and latitudinally and drift zonally. As a result, the detection longitudes of EPBs are slightly variable with altitude, latitude, and local time. It would be more reasonable to say that the whole longitude regions identified in Figures 2e and 2f were turbulent on the night of 22 September 2002 rather than to say that the ionosphere were turbulent only at the exact locations of plasma depletions identified in Figures 2g–2j.
 Combining the 14–15 orbits of data for each day and binning them into 10° longitude bins, we create a daily EPB occurrence map. We identify the occurrence of EPBs when the number of data points for σ > 0.3% within the 10° longitude bin is greater than 50. This number is determined from the visual comparison of the daily EPB occurrence maps with the density plots. Figure 3 shows the EPB occurrence maps on the night of 22–23 September 2002 produced using the new method (Figures 3a and 3b) and conventional method (Figures 3c and 3d). The black dots in Figure 3a show the satellite paths within ±20° dip latitudes during 1900–0600 LT. Detection of EPBs using the new method (100-s data) is indicated by red symbols. Figure 3b shows the EPB distribution map on that night determined using the new method. The red color indicates the occurrence of EPBs (100%), the white color indicates the absence of EPBs (0%), and the black color indicates that the occurrence of EPBs is not determined. In Figure 3c, the occurrence of EPBs (red dots) is identified using 10-s data. The EPB occurrence probability determined using the conventional method (Figure 3d) shows large longitudinal fluctuation. The EPB occurrence probability is smaller than 50% in all longitude bins in the conventional method, although numerous EPBs were detected on that night. The new method identifies the occurrence of EPBs continuously in broad longitude regions except between 140°–60°W. The conventional method finds that EPBs are absent between 140°–60°W. In the new method, the occurrence of EPBs is undetermined between 120°–60°W since the observations were made early in the morning.
 The daily EPB occurrence maps are produced by analyzing all ROCSAT-1 data. On each day, the occurrence of EPBs (0: bubble absent, 1: bubble present, 2: undetermined) is recorded for every 1° longitude bin. Figure 4 shows the stack of the daily EPB occurrence maps during March 1999 to June 2004 for Kp ≤ 3+. The red color indicates the occurrence of EPBs, the white color indicates the absence of EPBs, and the black color indicates that the occurrence of EPBs is not determined. While many locations of the map remain “undetermined” (black), we can identify the repetition of similar EPB occurrence pattern year after year.
3. Global EPB Distribution
 The monthly EPB occurrence probability is calculated using the daily EPB occurrence maps. The occurrence probability at a certain longitude and in a month bin is obtained by dividing the number of the EPB detections (index 1) by the total number of determined cases (indices 0 and 1) within the bin. Figure 5 shows the variation in the EPB occurrence probability with longitude, month, and year. The monthly and longitudinal variations in the EPB occurrence percentage are very similar during the years of 1999–2002. While the EPB occurrence probability is reduced in 2003–2004 compared to that in previous years, its global morphology is not much different from that in earlier years. The decrease in the EPB occurrence probability from 2003 is attributed to a concomitant decrease in the solar flux.
 To examine the correlation between the EPB occurrence rate and the PRE, the mean EPB occurrence probability map is produced using data during 1999–2002 for Kp ≤ 3+. The measurements of the vertical ion velocity within ±5° dip latitudes at 1730–1930 LT during the same period were processed to produce the global PRE map. The occurrence time of the PRE slightly varies with season and longitude [e.g., Fejer et al., 2008] but the global PRE morphology does not change much by shifting the time interval forward or backward by 30 min. Figure 6 shows the global maps of EPB occurrence probability (middle) and PRE (bottom). The EPB occurrence probability is high in most months in the longitude range of 0°–40°E. The EPB occurrence probability is over 50% in March and September in most longitude regions. During May–August the EPB occurrence is severely suppressed in the longitude regions 110°–20°W and 60°–120°E. Near December the EPB occurrence probability is small except for longitudes of 90°W–30°E. The global EPB distribution morphology is similar to the results obtained from AE-E [Kil and Heelis, 1998; McClure et al., 1998; Hei et al., 2005] and ROCSAT-1 [Su et al., 2006]. However, the magnitude of the EPB occurrence probability is about 2∼3 times larger than that in previous studies of the in situ data. The magnitude of the EPB occurrence probability in Figure 6 is comparable to that obtained from ground observations [Makela et al., 2004; Nishioka et al., 2008]. The EPB distribution morphology is almost identical to that produced using DMSP data during 1999–2002 [Burke et al., 2004a; Gentile et al., 2006]. The DMSP-derived maximum EPB occurrence probability during solar maximum is about 50%. The smaller EPB occurrence probability observed by DMSP than that observed by ROCSAT-1 can be attributed to the difference in the two satellite heights. Stolle et al.  produced a similar EPB distribution map using the measurement of the magnetic field from the Challenging Minisatellite Payload (CHAMP) satellite during 2001–2004. The EPB occurrence probability obtained from the CHAMP data is smaller than that obtained from the ROCSAT-1 analysis, although the altitude of CHAMP satellite (450 km) is lower than that of ROCSAT-1 (600 km). The EPB occurrence probability calculated by Stolle et al.  is the mean occurrence probability during 1800–0400 LT. As was pointed out earlier, EPBs are normally detected after 1900 LT. Since EPBs are rarely detected before 1900 LT and decay after their creation, inclusion of data from a broad local time range might have reduced the mean EPB occurrence probability deduced from the CHAMP data.
 The morphologies of the EPB and PRE distributions show a close similarity. The magnitude of the mean PRE during northern summer is small in most longitudes, especially in the longitude ranges 90°–30°W and 60°–120°E. Near the equinox months the PRE is particularly pronounced. The PRE reaches large values in all seasons in the longitude range 0°–30°E. The characteristics of the PRE distribution coincide with the characteristics of the EPB distribution.
Figure 7 shows the relationship between the EPB occurrence probability and the magnitude of the PRE obtained using the same data sets presented in Figure 6. The EPB occurrence probability is seen to linearly increase with an increase in the magnitude of the PRE. The correlation coefficient is 0.75. The occurrence probability varies from 0% to 100%. In the previous studies, the variation of the EPB occurrence probability was less than 40% [Li et al., 2008; Su et al., 2008]. The EPB occurrence probability is highly variable at a given PRE value. For a PRE value of 30 m s−1, for example, the EPB occurrence probability varies from about 40% to 100%. In other words, the generation of EPBs cannot be explained solely by the effect of the PRE. The PRE may provide a preferred condition for the development of R-T instability but the triggering of the R-T instability by some other means such as gravity waves or LSWS may be required for the generation of EPBs. We also need to consider whether the mean velocity observed from 1730 to 1930 LT can be representative of the PRE. The effects of some other PRE characteristics such as onset time, duration, and peak velocity, in addition to the mean vertical velocity in the evening, would be useful parameters to investigate. We note that EPBs can be produced quite a long time after the PRE. EPBs produced after midnight would not have any connection with the PRE. The uncertainty in the creation times of EPBs may also be one of the sources of the variability. The PRE-EPB relationship needs to be further improved by considering the effect of those factors.
 The EPB occurrence probability obtained from low-inclination orbit satellites is dependent on the data sampling and EPB counting method. As a result, the EPB occurrence probabilities obtained to date from the AE-E and ROCSAT-1 satellites have been significantly smaller than those obtained from ground observations. We have developed a new EPB detection method to resolve the problems in the analysis of satellite data. The occurrence morphology of EPBs obtained by applying the new EPB detection method to the ROCSAT-1 data is very similar to that reported in previous studies. However, our new EPB occurrence probability is a few times greater than the EPB occurrence probability obtained by applying the conventional method. The new EPB occurrence statistics show good agreement with ground observations. Our new EPB detection method can be directly applicable to the ion density measurements from the Coupled Ion Neutral Dynamic Investigation (CINDI) instrument on board the Communication/Navigation Outage Forecasting System (C/NOFS) satellite launched in April 2008.
 The similarity in the global morphologies of the EPB occurrence probability and the mean vertical velocity at 1730–1930 LT supports the notion that the PRE is an important factor in the generation of EPBs. The longitudinal EPB distribution does not show the wave 4 or wave 3 structure that has been verified in the longitudinal plasma density in recent studies [Sagawa et al., 2005; England et al., 2006a, 2006b; Immel et al., 2006; Kil et al., 2007, 2008; Lin et al., 2007; Lühr et al., 2007; Oh et al., 2008; Scherliess et al., 2008]. Our results show the presence of a linear relationship between the EPB occurrence probability and the magnitude of the PRE. However, the variation range of the EPB occurrence probability at a given PRE value is quite large. A significant number of cases shows the absence of EPBs for intense PRE. In additional to the provision of preferred growth condition of the R-T instability, the presence of a precursor such as LSWS or other initial seed perturbations need to be considered to explain the generation of EPBs.
 H. Kil and L. J. Paxton acknowledge support from NASA grant NNX08AQ12G. H. Kil thanks S. -Y. Su for providing the ROCSAT-1 data.
 Wolfgang Baumjohann thanks the reviewers for their assistance in evaluating this paper.